| As the digital technology develops, a large amount of visual data (images and videos) are captured and shared every day in the social media. It is the fact that most data suffer from lack of tags or labels for various reasons, e.g. high labeling cost, etc. To represent the unlabeled visual data, extensive efforts have been made in both industry and academia. For instance, Apple launched their face clustering application in iPhone iOS-10 system to help users organize their photos by identity. However, it suffers from many criticisms, including poor clustering performance with occluded faces, non-frontal faces etc. The fundamental problem behind this challenging problem is, how to robustly represent the photos under different variations, such as head poses, lighting conditions, occlusions, or even larger corruptions. In this dissertation, we focus on solving robust visual data representation problems using unsupervised subspace learning (clustering) techniques.;According to the number of input data views (modalities), unsupervised subspace clustering methods are usually divided into two categories, i.e. single-view subspace clustering and multi-view subspace clustering. In this dissertation, both cases are discussed. Specifically, in single-view subspace clustering (Part 1), we propose a novel graph-based method, ESSB: Ensemble Subspace Segmentation under Block-wise constraints, which unifies least squares regression and locality preserving graph regularizer into an ensemble learning framework. The "divide-and-conquer'' strategy is applied to features, resulting in an efficient framework to handle the high-dimensional data. For the large-scale data, we propose a Fast Regression Coding (FRC) scheme to optimize regression codes, and simultaneously train a non-linear function to approximate the codes. By using FRC, we develop an efficient Regression Coding Clustering (RCC) framework to solve the large-scale clustering problem, consisting of sampling, FRC and clustering. Besides, we provide a theorem to guarantee that the non-linear function has a first-order approximation ability and a group effect. The theorem manifests that the codes are easily used to construct a dividable similarity graph.;In multi-view subspace clustering (Part 2), we firstly present a Deep Matrix Factorization (DMF) framework for traditional multi-view clustering problem, where semi-nonnegative matrix factorization is adopted to learn the hierarchical semantics of multi-view data in a layer-wise fashion. To maximize the mutual information from each view, we enforce the non-negative representation of each view in the final layer to be the same. Beyond this, a more challenging multi-view scenario, i.e. missing view information, is considered. This usually happens in practice, e.g. when one sensor malfunctions. To address this, a novel robust graph regularized method is proposed to handle the incomplete data by projecting the original-and-incomplete data to a learned-and-complete latent space. Finally, as a brand-new application, we study the outlier detection problem. Contrary to single-view outlier detection, we define the types of outliers as attribute-type and class-type in multi-view setting. By representing the multi-view data with latent coefficients and sample-specific errors, the proposed consensus regularized multi-view outlier detection method is able to detect both types of outliers simultaneously. |
